MCP server
A voice AI platform with an MCP server
Yes. VoiceDock is a voice AI platform with a hosted MCP server, so you can build and manage voice agents from Claude, ChatGPT, Cursor or another MCP client. Add https://mcp.hmsovereign.com/mcp to your client, sign in with your VoiceDock account, and every endpoint of the VoiceDock API becomes a tool: create and change assistants, put them on phone numbers, start outbound calls, read transcripts and check usage. ElevenLabs, Vapi and Retell AI run MCP servers too; the comparison further down shows how they differ.
mcp.hmsovereign.com/mcp · sign in with OAuth 2.1, no API key to paste · one tool per API endpoint
What you can do
What a connected AI assistant can do in VoiceDock
The server is built from the VoiceDock API, so it covers the same ground. You ask in plain language; the assistant picks the tool.
Build and change voice agents
Create an assistant from a short description, then change its prompt, voice, model or tools. Ask for the current configuration first, so the assistant edits what is there and not what it remembers. The model you choose also decides where the audio of a call is processed.
Put agents on numbers and place calls
Attach an assistant to a phone number, start an outbound call to test it, or set up a campaign of outbound calls. Phone numbers in the docs.
Read what happened on a call
List calls with filters, open one for its transcript and analysis, and ask for a summary of the week. That is often the quickest way to find the turn where a prompt needs work. Calls in the docs.
How to connect
Connect your AI assistant in four steps
The same steps for every client that supports remote MCP servers with OAuth.
- 01
Add the server URL to your client
In Claude Code:claude mcp add --transport http voicedock https://mcp.hmsovereign.com/mcp. In Claude Desktop: Customize → Connectors → Add custom connector, then paste the URL. In Cursor: add the URL in the MCP settings. Setup per client. - 02
Sign in with your VoiceDock account
Your client opens a browser. You log in and approve the access on the consent screen. There is no API key to copy, and you can revoke the connection from your client at any time. - 03
Choose the organisation
With one organisation, every call runs there. With more than one, the assistant passesorganization_idon every call, reads included; without it, nothing is read or changed and you get the list of your organisations back. - 04
Ask for what you want
“Create an assistant called Support with a friendly greeting.” “List this week's calls and summarise the outcomes.” “Show my usage for the last 30 days.” The server can create, change and delete, so keep confirmation before tool calls switched on in your client. Want to see a whole build done this way? Read how to build an AI receptionist platform with Claude Code.
Compared
Four voice AI platforms with an MCP server
ElevenLabs, Vapi and Retell AI also offer an MCP server for managing voice agents from an AI assistant. The table below is taken from each vendor's own page, read on 31 October 2026: ElevenLabs, Vapi, Retell AI and VoiceDock. For the full picture beyond MCP, see our pages on Vapi and Retell AI.
| VoiceDock | ElevenLabs | Vapi | Retell AI | |
|---|---|---|---|---|
| Server | Hosted, at mcp.hmsovereign.com/mcp | Hosted, at api.elevenlabs.io/v1/mcp, with separate URLs for its EU, India and Singapore data residency environments | Hosted, at mcp.vapi.ai/mcp and an SSE endpoint, or run locally with npx | Hosted, at mcp.retellai.com |
| Sign-in | OAuth 2.1 with your VoiceDock account; an organisation API key still works for scripts and CI | OAuth with your ElevenLabs account | API key as a bearer token | API key as a bearer token; the separate connector in Claude's directory signs in with OAuth |
| Tools | One tool per API endpoint, generated from the API specification | Agent management plus conversations, tests, branches, knowledge base, cost estimates and speech generation | Ten core tools for assistants, calls (including scheduled calls), phone numbers and tools | Three tools that look up any API endpoint, read its schema and call it |
| Setup documented for | Claude Desktop, Claude Code, Cursor and any client with Streamable HTTP and OAuth | Claude Desktop, where it is listed in the connector directory, and other OAuth clients | Claude Desktop, the OpenAI Responses API, and Cursor, Windsurf and VS Code through the Vapi CLI | Cursor, Claude Desktop, Claude Code and Codex, plus the connector in Claude's directory |
| More than one account | Every organisation your account belongs to, chosen per call | The workspace you sign in with | The account of the API key | The account of the API key; in the Claude connector, the workspace you pick |
Fair to say
Where another MCP server may suit you better
Each of these is a real advantage, and worth weighing before you choose.
ElevenLabs is one click away in Claude
Its server is listed in the connector directory of Claude Desktop, so you connect it without pasting a URL. Its tools also go further into agent operations: tests, branches and drafts, and a cost estimate before you change a model.
ElevenLabs has data residency endpoints
Workspaces in its EU, India or Singapore environments connect through their own server URL. On VoiceDock, orchestration, dashboard, API and storage run on EU infrastructure for every account.
Vapi runs locally and configures your editor
Besides the hosted server, Vapi publishes a local server you start with npx, and its CLI writes the MCP configuration for Cursor, Windsurf or VS Code in one command.
Retell keeps the tool list short
Retell offers three tools that look up and call any endpoint, which leaves more room in your assistant's context. VoiceDock lists one tool per endpoint: a longer list, but the assistant sees every parameter without an extra lookup. Retell's connector in Claude's directory adds OAuth sign-in and widgets inside the chat, for agents built on a single prompt.
Related
Where to go next
Developers
The REST API, the TypeScript SDK and signed webhooks behind the MCP server.
Connecting your systems
How an assistant reaches your CRM or ERP during a call: webhook tools, your own MCP server and webhooks.
AI receptionist
What a platform for an AI phone receptionist needs, and how a call runs on VoiceDock.
FAQ
Questions about the VoiceDock MCP server
Is there a voice AI platform with an MCP server so I can build and manage voice agents from Claude or ChatGPT?
Yes. VoiceDock runs a hosted MCP server at https://mcp.hmsovereign.com/mcp. Connect Claude, ChatGPT, Cursor or another MCP client, sign in with your VoiceDock account, and you can create assistants, put them on numbers, place calls and read transcripts in plain language. ElevenLabs, Vapi and Retell AI offer MCP servers as well; the comparison on this page shows how they differ.
Does it work with ChatGPT?
Our setup guide covers Claude Desktop, Claude Code and Cursor. ChatGPT lets you add a custom MCP server by its URL, as OpenAI's documentation describes; whether you can do that depends on your account and workspace policies. Add https://mcp.hmsovereign.com/mcp and sign in with your VoiceDock account when you are asked to.
Which tools does the VoiceDock MCP server offer?
One tool per endpoint of the VoiceDock API, generated from its specification: assistants, phone numbers, calls, campaigns, voices, usage and everything else in the API reference, plus a tool that lists your organisations. A new endpoint becomes a tool once we have restarted the server; there is nothing to configure on your side.
Do I need an API key?
No. You sign in with your VoiceDock account through OAuth 2.1 and approve the access on a consent screen. For a server or CI job that cannot sign in, an organisation API key still works as a bearer token. Both options in the docs.
Is it safe to let an AI assistant change my voice agents?
Treat it like giving someone access to your account, because the server can create, change and delete. Approve only connections you started yourself, keep confirmation before tool calls switched on in your client, and know that membership is checked on every call: removing someone from an organisation ends their access through the MCP server straight away.
Can a voice agent use my own MCP server during a call?
Yes, and that is the other direction. You add your MCP server to your organisation with the headers it needs, such as an API key, and list in allowed_tools exactly which of its tools the assistant may use. There is no sign-in flow on that side: you pass a token as a header. MCP servers as tools explains the setup, and designing tool calls for a live phone line what to watch for.
What does it cost?
The MCP server has no price of its own. Calls you start through it are billed like any other: € 0,07 per minute for orchestration, model usage at cost or on your own keys, and telephony per leg. Our pricing page lists every item.
Manage your voice agents from your AI assistant
Create an account and add the server to Claude, ChatGPT or Cursor, or plan a call if you want to build your voice agents with us. The developer page covers the API and SDK behind it.